Understanding small ORF diversity through a comprehensive transcription feature classification

Abstract

Small open reading frames (small ORFs/sORFs/smORFs) are potentially coding sequences smaller than 100 codons that have historically been considered junk DNA by gene prediction software and in annotation screening; however, the advent of next-generation sequencing has contributed to the deeper investigation of junk DNA regions and their transcription products, resulting in the emergence of smORFs as a new focus of interest in systems biology. Several smORF peptides were recently reported in non-canonical mRNAs as new players in numerous biological contexts; however, their relevance is still overlooked in coding potential analysis. Hence, this review proposes a smORF classification based on transcriptional features, discussing the most promising approaches to investigate smORFs based on their different characteristics. First, smORFs were divided into non-expressed (intergenic) and expressed (genic) smORFs. Second, genic smORFs were classified as smORFs located in non-coding RNAs (ncRNAs) or canonical mRNAs. Finally, smORFs in ncRNAs were further subdivided into sequences located in small or long RNAs, whereas smORFs located in canonical mRNAs were subdivided into several specific classes depending on their localization along the gene. We hope that this review provides new insights into large-scale annotations and reinforces the role of smORFs as essential components of a hidden coding DNA world.

1. Introduction

The big data era promoted by next-generation sequencing (NGS) has irreversibly changed genetics and systems biology. The amount of biological data available to scientists has been rapidly increasing during the last 15 years,¹ which makes the intersection between experimental sciences and computational biology even more imperative to overcome new scientific barriers.

One of the most important achievements in NGS technology has been the development of deep transcriptome sequencing approaches, such as RNA-seq and ribosome profiling, which have greatly improved proteomic, peptidomic and phenotypic analyses.^2–4 New molecular components have thus been discovered,⁵ especially in hidden proteomes expressed from coding small open reading frames (small ORFs/sORFs/smORFs).⁶

Generally, open reading frames (ORFs) are defined as nucleotide sequences between a translation start codon and the nearest in-frame stop codon (Fig. 1).⁷ SmORFs differ from other ORFs in size, which typically range from the lower theoretical limit of two codons⁸ to 100 codons⁹; however, upper thresholds from 150 to 250 codons have rarely been proposed in the literature.^8,10,11 In prokaryotes, a 50-codon size limit is normally accepted.¹²

Figure 1.

smORF peptide biosynthesis. (A) smORF transcription, translation and cellular/extracellular trafficking. smORF peptide biosynthesis occurs directly via ribosome translation after smORF gene transcription. smORF peptides can play several roles inside and outside the cell. RNA polymerase in the nucleus is shown in green; ribosomes in the cytoplasm are shown in red; and smORF peptides in the cytoplasm are shown as blue winding lines. (B) Schematic representation of a hypothetical ORF. The illustrated ORF is a smORF within the first of the three RNA frames (Frame 1). The smORF is highlighted in bold font; the start codon is shown in green; the stop codon is shown in red; and the remaining codons are shown in blue. Above the smORF codons are their corresponding one-letter-code amino acids, encoding a hypothetical 11 amino acid smORF peptide.

SmORFs have historically been dismissed in annotation screenings due to methodological challenges. Pioneering smORF studies performed before the NGS era (e.g., Refs^13–17) were even more challenging without deep sequencing data, but greatly contributed to the acceptance of the field when new sequencing technologies were developed. Currently, gene prediction algorithms and genome annotations usually overlook smORFs owing to their low statistical coding potential,^18,19 which has been justified by the fact that millions of non-functional smORFs occur stochastically in genomes due to their small size. Thus, the computational prediction and experimental analysis of coding smORFs are challenging tasks, akin to searching for a needle in the haystack^10,20; however, several biologically relevant smORF peptides have been discovered throughout the three domains of life (study in Archaea²¹; Eukarya review²²; Bacteria review¹²) and viruses.²³ SmORFs are also significant targets of biomedical studies.^24–31 Moreover, recent discussions have revealed that smORFs are important precursors of de novo protein-coding gene birth.^32,33 Thus, although smORFs were previously hidden among the so-called ‘junk DNA’, their study is now an emerging field.

Our new classification scheme comprises 12 smORF classes and subclasses based on transcription features. Some of the classes proposed have been covered in previous reviews and screening reports during recent years, under different perspectives.^34–39 Here, we contribute to this knowledge by offering a ‘divide and conquer’ perspective to discuss the most promising approaches to study each smORF class. By providing a detailed explanation of the smORF classes and organizing a comprehensive scheme, we note how the features of each smORF group influence the coding smORF detection. We also discuss the molecular findings that support the coding potential of each class and the future prospects for the smORF field. We expect that this classification scheme may direct further studies and discoveries of new coding smORFs by offering a comprehensive landscape of smORF diversity and their detection challenges.

First, smORFs are divided into two general groups, namely expressed and non-expressed. Non-expressed smORFs are also known as intergenic smORFs (Fig. 2A).³⁴ Second, the expressed smORFs are subdivided into two categories: smORFs located in canonical mRNAs or smORFs located in transcripts with non-coding characteristics (Fig. 2A); in the second case, transcripts may evolve dual functional roles. Finally, we subdivided the two expressed smORF groups into several other classes according to their transcriptional features (Fig. 2B). Importantly, all proposed classes may contain random and non-coding smORFs, with the exception of the group where the smORFs are the reference CDSs (coding DNA sequences) along mRNAs. Throughout the text we focus on the study of the coding smORFs of each class.

Figure 2.

Proposed smORF classification. (A) smORF classes and their representative locations. Hundreds of thousands of smORFs in the genome are non-expressed and are therefore classified as intergenic smORFs (green box). Expressed smORFs are classified as genic smORFs (green box) and are subdivided into smORFs located in non-coding RNAs (ncRNAs) (red box) and smORFs located in canonical mRNAs (blue box). Different types of ncRNAs and canonical mRNAs with their respective classes of smORFs are represented in red and blue boxes. Red tracks represent intergenic smORFs; blue tracks represent genic smORFs; yellow tracks represent large ORFs. (B) smORF classification chain. SmORF classes can be organized into groups and subgroups defined by transcriptional features.

2. Intergenic smORFs

Intergenic smORFs are not generally functional or transcribed; they constitute the great majority of smORFs in the genome and originate randomly via the simple arrangement and rearrangement of nucleotides, causing them to be classified as junk DNA.³⁴

2.1. Strategic insights into the investigation of intergenic smORFs

Analysis based on sequence similarity indicated that thousands of theoretically non-expressed smORFs are evolutionarily conserved, which suggests coding potential.⁴⁰ Importantly, most coding smORFs are overlooked in genome annotation screenings and are usually classified as intergenic DNA stretches. Thus, strategies to uncover new intergenic smORFs must include smORF detection in intergenic regions using bioinformatic tools, such as the getORF program provided by EMBOSS,⁴¹ followed by filters of evolutionary conservation and comparison to RNA sequence databases to verify if the sequence is expressed (a similar approach was performed in Ladoukakis et al.⁴²). Then, an analysis of differences in expression under stress conditions is a useful approach since recent evidence from bacteria suggests that intergenic smORFs can be expressed under osmotic stress, lower temperatures and specific growth conditions, indicating that intergenic smORFs are a potential source of peptides with condition-dependent expression.⁴³

3. smORFs in small RNAs

Small RNAs are transcripts smaller than ∼200 − 300 nucleotides.^44–46 The regulatory roles of small RNAs are widely studied in all domains of life,^47–49 as small RNAs are involved in cellular responses to biotic and abiotic stresses^50,51; however, the coding capacity of small RNAs has been neglected, and reports of functional smORFs within this type of transcript exist but are scarce.

3.1. Strategic insights into the investigation of smORFs in small RNAs

The most promising source of smORFs in small RNAs are prokaryotic dual-function transcripts, as previously reported.^52–54 Strictly coding small RNAs are an undescribed phenomenon, but their existence is theoretically possible. The coding capacity of eukaryotic small RNAs is still unexplored and requires investigation.

In prokaryotes, the widespread distribution of potentially coding small RNAs has been confirmed by computational predictive approaches. For instance, it was recently shown that at least 0.5% of all small RNAs in 14 bacterial species contain smORFs under purifying selection, and the proportion reaches ∼20% in some taxa.⁵⁵ Moreover, mass spectrometry and ribosome profiling database analyses have confirmed that dozens of these smORFs in small RNAs are in fact translated.⁵⁵ Thus, a bioinformatic approach integrating small RNA identification^56,57 followed by smORF detection pipelines based on evolutionary conservation and ribosome profiling footprints^58–61 must be considered to detect new coding small RNAs in large-scale screening. In addition, the expression under different stress conditions is certainly an important factor to be analysed, because at least 40 putative dual-function small RNAs encode smORFs smaller than 30 codons in Methanosarcina mazei (Archaea) and their expression is modulated by different levels of nitrogen availability.⁶² Importantly, the experimental detection of small RNAs is highly dependent on the RNA isolation method because small RNAs can be lost during precipitation steps due to their size. Column-based or acrylamide gel isolation approaches should be preferentially applied for the isolation of potential smORF-containing small RNAs.

Because evidence suggests that most coding small RNAs are dual-functional (examples in the section below), another interesting strategy is the search for smORF sequences in known regulatory small RNAs using bioinformatic tools, such as the Expasy Translate Tool.⁶³ Then, each hypothetical smORF peptide identified should be submitted to tBLASTn analysis⁶⁴ against related species databases applying non-stringent parameters for smORF detection.⁶⁵ Finally, putative smORFs could be indicated by evolutionary conservation. The use of hypothetical peptide sequences in BLAST searches avoids false negative results caused by synonymous modifications in small coding sequences. This strategy can be adapted for any smORF class.

3.2. Examples of smORF peptides from small RNAs

One of the first described dual-function small RNAs was SgrS, which is transcribed under glucose-phosphate stress in Escherichia coli.⁶⁶ Interestingly, both the SgrS transcript and its smORF peptide play roles in glucose flux, but in different pathways. The SgrS transcript downregulates an important glucose transporter by base pairing with its mRNA, and SgrS also encodes a 43 amino acid peptide, sgrT, that modulates the influx of glucose by inhibiting its transporters.⁶⁶

The functional analysis of another small RNA, SR1, in Bacillus subtilis showed that the coding and non-coding activities of this small RNA are distinct.⁶⁷ The regulatory role of SR1 involves the inhibition of the translation of an important transcription activator via nucleotide pairing. The coding activity of SR1 is mediated by a 39 amino acid peptide, SR1P, encoded by the transcript. SR1P binds to the glycolytic enzyme GapA, triggering the formation of a complex with RNAse J1 and thereby increasing the affinity of RNAse for its substrates.⁶⁷ In addition, SR1P-GapA binding promotes the stabilization of gapA operon mRNA.⁶⁷

In Staphylococcus aureus, the RNAIII transcript regulates the translation and/or stability of virulence factors, cell wall metabolism enzymes and transcription factor mRNAs (reviewed in Bronesky et al.⁵²). Moreover, RNAIII encodes the smORF peptide δ-haemolysin,⁶⁸ also known as hld (26 amino acids length), which can lyse red blood cells, trigger membrane disorders, and exert antimicrobial activities (reviewed in Verdon et al.⁶⁹).

Mass spectrometry analysis of the archaeal species M. mazei undergoing cell growth under different stress conditions identified three smORF peptides between 23 and 61 amino acids in length.²¹ The identified smORF peptides exhibit high conservation among Methanosarcina species as well as highly conserved secondary structures of their small RNAs. Based on the concentration of the peptides during nitrogen restriction stress, oligopeptide 36 (61 amino acids) might modulate an essential protein associated with nitrogen metabolism.²¹

4. smORFs in long non-coding RNAs

By definition, long non-coding RNAs (lncRNAs) are untranslated RNA molecules longer than ∼200 nucleotides.³⁵ Typical lncRNAs are expressed at low levels and are non-conserved. On the other hand, lncRNA expression is associated with biological processes such as differentiation, proliferation, embryonic development, cancer, apoptosis and stress responses (reviewed in Chekulaeva and Rajewsky⁹ and Perry and Ulitsky⁷⁰).

The coding potential of lncRNAs is still disregarded owing to certain lncRNA characteristics, including (i) the absence of major ORFs⁷¹; (ii) degeneration frequency similar to that of introns and much higher than that of exons, which disfavours new coding ORF fixation^72,73 and (iii) when translated, the instability and rapid degradation of lncRNA peptides, suggesting that the translation of these peptides is a random and neutral process.⁷⁴ In contrast, the effective translation of smORFs located in lncRNAs has been described, and these translated smORFs exhibit sequence, structural and functional conservation.^24,31,75,76 Currently, smORF peptides translated from lncRNAs are considered to represent a new frontier in biomedical studies focussed on new biomarkers and molecular targets in cancer.²⁹

Our classification scheme indicates that smORFs discovered in strictly ‘coding lncRNAs’ should be classified as small reference coding sequences (small refCDSs) when their transcripts are reannotated as mRNAs (Fig. 3). Therefore, small refCDS functions will be discussed in its corresponding section. In the case of coding circular RNAs (circRNAs) lacking non-coding functional annotation, we still classify their coding smORFs as smORFs in lncRNAs, because the field is recent and the molecular niche of circular transcripts is poorly known, but recent evidence points to regulatory effects without translation.^77–79

Figure 3.

General scheme of the smORF classification of misannotated non-coding transcripts. Strictly coding lncRNAs are reclassified as smORF located in mRNAs, while ncRNAs showing both coding and regulatory roles are reannotated as dual-function transcripts. Three smORF classes can be identified in ncRNAs: small refCDSs, smORFs in lncRNAs (dual functional) and smORFs in small RNAs (generally dual functional). Red tracks represent coding smORFs.

4.1. Strategic insights into the investigation of smORFs in lncRNAs

Among all smORF classes, lncRNAs are the most promising source of smORF discoveries. For instance, ∼98% of annotated lncRNAs in Metazoa contain at least one unannotated smORF.³⁴ Studies in Arabidopsis thaliana show that thousands of smORFs in lncRNAs display evidence of translation, as confirmed by Ribo-Seq analysis.⁸⁰ Another clue is provided by the wide gene coverage of smORFs, because ∼70% of human genes are expressed as lncRNAs.⁸¹ These data show that a plethora of ‘coding lncRNAs’ are still misannotated or even overlooked.

In this context, some features should be considered in annotation screenings searching for smORFs in lncRNAs. For instance, the amino acid usage of theoretical unannotated smORF peptides in lncRNAs differs from that of canonical proteins, although it is not random,³⁴ suggesting a new biological pattern that should not be dismissed. Another important feature is that misannotated lncRNAs can exhibit typical mRNA structures or non-canonical characteristics.^82,83 In the second case, in addition to the absence of polyadenylation sites, Kozak consensus sequences and 5′ caps,^84,85 in some cases these transcripts exhibit a circular structure.^86,87 Furthermore, lncRNAs are usually non-conserved, although their smORFs can still show signs of positive selection at amino acid sequences.⁶¹ However, even in the absence of orthologues, species-specific smORFs can be functional via orphan gene generation.⁸⁸

Bioinformatic prediction approaches are a promising strategy to detect putative smORFs based on the aforementioned features, examples include machine learning programs⁸⁹ and codon usage comparisons between random and functional smORF peptides. Importantly, the diversity of lncRNA types enables several approaches to investigate their coding potential. For instance, some types of housekeeping lncRNAs can also encode smORF peptides (Fig. 3), making them dual-function transcripts, as described for rRNAs,⁹⁰ and pri-miRNAs.⁹¹ The translation of smORFs in lncRNAs might even include antisense and intronic transcripts.⁹² Thus, smORF detection performed on known functional lncRNAs followed by comparative analysis using BLAST is also an interesting strategy.

After the selection of potentially coding smORFs, a promising experimental approach is the development of precise CRISPR-Cas9 genome editing, such as precise deletions or point mutations, on conserved smORFs in lncRNAs differentially expressed in important biological contexts, such as cancer. Recent yeast and fruit fly studies used similar strategies.^93,94

4.2. Examples of smORF peptides from lncRNAs

Several smORF peptides have been recently discovered in housekeeping lncRNAs. For example, the pri-miR171b and pri-miR165a transcripts, which are miRNA precursors, encode two peptides of 18 and 21 amino acids, respectively. Both peptides exhibit cis-acting regulatory functions by increasing the accumulation of their correlated miRNAs, thereby indirectly promoting the negative modulation of their targets associated with root development in Medicago truncatula and A. thaliana.⁹¹

Another example of a dual-function transcript is mammalian mitochondrial 12S rRNA, which encodes the smORF peptide MOTS-c (16 amino acids), a regulator of metabolic homeostasis in the nucleus.^90,95 In mice, MOTS-c treatment prevents insulin resistance and diet-induced obesity and reverses age-related insulin resistance in muscles.^90,95 Additionally, MOTC-c is involved in cold stress defence by increasing adipose thermogenesis.⁹⁶

The smORF peptide humanin (24 amino acids), encoded by human mitochondrial 16S rRNA,¹³ was identified as a new cDNA involved in Alzheimer’s disease. Humanin interacts with the apoptosis regulator Bax (BCL-2-associated X protein), preventing its activation and, thus, cell death via apoptosis.¹³

For many years, circRNAs have been considered atypical products in cells; however, it was recently reported that these molecules are stable and are generated via a mechanism known as back-splicing.^77,97 circRNAs might act as miRNA and protein sequesterers and may be involved in the splicing regulation of RNA polymerase II-mediated transcription (reviewed in Refs^77–79) Moreover, circRNAs have been shown to be involved in several pathologies, such as diabetes, neurological diseases and cancer.⁷⁸ Interestingly, circRNAs lack optimal translation sequences, such as the 5′ end 7-methylguanosine (m⁷G) cap structure and 3′ poly(A) tail.⁹⁷ This new class of regulatory RNAs can also be translated into conserved smORF peptides.^{85–87,98,99}

An example of a coding circRNA is circMbl1, which encodes a peptide of ∼10 kDa that moves to synapses in response to starvation and FOXO expression, suggesting that smORF peptides encoded by circRNAs may be involved in the mechanisms of neuronal communication.⁸⁵ Another example of a coding circRNA is the circular form of (Long Intergenic Noncoding RNA p53-Induced Transcript (LINC-PINT), a glioblastoma suppressor,⁸⁷ which encodes an 87 amino acid smORF peptide that directly interacts with Polymerase Associated Factor complex (PAF1c), thereby inhibiting transcription elongation of multiple oncogenes. Interestingly, its expression is lower in glioblastoma cells than in normal tissues.⁸⁷ Finally, the circPPP1R12A circRNA gene triggers tumour pathogenesis and metastasis in colon cancer by activating the Hippo-YAP signalling pathway. Interestingly, this circRNA gene contains a smORF encoding a 73 amino acid peptide.⁹⁹ All of these findings suggest that circRNAs are the newest reservoir of smORFs to be explored.

5. smORFs as reference CDSs

A reference CDS is the main or unique coding ORF of an mRNA, which can be flanked by translatable alternative ORFs.¹⁰⁰ All reference CDSs smaller than 100 codons are defined as small refCDSs in this classification. A strong criterion for the annotation of the coding capacity of a transcript is related to ORF size.¹⁰¹ Thus, size restriction during automatic annotation was probably one of the reasons for the mis-annotation of important small refCDS transcripts as long non-coding RNAs in past years.^{14,24,25,30,31,91,102–108}

5.1. Strategic insights into the investigation of smORFs as reference CDSs

Studies based on ribosome profiling and mass spectrometry are promising approaches because they have indicated the widespread translation of potential small refCDSs, but often at low expression rates,^{37,58,60,109–111} which possibly indicates the coding potential immaturity of the transcripts.³³ Moreover, some eukaryotic mRNAs exhibit more than one small refCDS, constituting polycistronic transcripts (Fig. 4), which might even contain conserved and duplicated smORFs.^24,65,112 Thus, a strategy based on a crosslink between ribosome profiling and mass spectrometry can indicate which smORFs are translated, even in polycistronic transcripts. Additionally, analysis of conservation at the amino acid level among different species coupled to transcript and peptide localization techniques (e.g. in situ hybridization and antibody staining, respectively) are promising approaches to identify biologically relevant small refCDSs. Knockdown experiments using RNA interference (RNAi) or loss-of-function via CRISPR-Cas9 are also important techniques if available in selected species; however, RNAi results are not conclusive when applied to polycistronic transcripts because they do not reveal which smORF is responsible for the studied phenotype.

Figure 4.

Comparison between large ORF mRNAs and smORF transcripts, which can occur in polycistronic arrangements. The yellow track represents large reference CDS; blue tracks represent small reference CDSs (coding smORFs). The lower panel represents a polycistronic mRNA containing three smORFs (blue tracks).

5.2. Examples of small refCDS peptides

One of the most emblematic examples of misannotated small refCDS transcripts is toddler/apela/ELABELA/ende, whose coding potential has been widely discussed (reviewed in Pauli et al.¹¹³) The toddler/apela/ELABELA/ende is a 55 amino acid peptide highly conserved among vertebrates that acts as an embryonic signal, binds to apelin receptors and triggers mesendodermal cell migration during gastrulation.⁷⁶

Another classic example is the polycistronic gene pri/mlpt/tal (polished-rice/mille-pattes/tarsal-less, respectively), first described in T. castaneum as a gap gene. Pri/mlpt/tal is highly conserved among Pancrustacea and encodes two to five duplicated peptides (10 − 30 amino acids) containing the same LDPTGXY motif.^65,112Pri/mlpt/tal is a well-known smORF gene, and its characterization has strongly contributed to the acceptance of smORFs as new developmental players in animals.²² Mutations and knockdown of pri/mlpt/tal promote lethal embryonic phenotypes that differ between species and among biological processes, such as the modification of epidermal structures,^106,114 changes in the number of locomotor appendages ^65,115,116 and tarsal deformations.^75,117 The main mechanism of action of pri/mlpt/tal smORF peptides is to trigger the truncation of the transcription factor Ovo/Shavenbaby N-terminus, thereby converting it from a repressor to an activator.¹¹⁸

In the Hemiptera order (bedbugs and aphids), our group identified a new smORF within the polycistronic gene pri/mlpt/tal, named smHemiptera.¹¹⁶ SmHemiptera consists of ∼80 codons in Rhodnius prolixus and exhibits a GHR(Y/N)WMTHLPLSRP region shared among all derived hemipterans whose pri/mlpt/tal sequence is available.¹¹⁶ In addition, smHemiptera contains two large introns,¹¹⁶ which is an uncommon pattern in genes encoding small proteins.¹¹⁹

Two transmembrane smORF peptides, sarcolamban A and B (28 and 29 amino acids, respectively), were discovered in a misannotated non-coding transcript of Drosophila melanogaster.²⁴ Sarcolamban peptides evolved 550 million years ago and contain orthologues with conserved roles in humans (phospholamban and sarcolipin).²⁴ These peptides are involved in calcium transport regulation in the sarcoplasmic reticulum and act specifically during myocardial contraction; as a result, several studies on cardiac arrhythmia target sarcolamban peptides and their orthologues.²⁴ Moreover, two smORF peptides from the same family, myoregulin (46 amino acids)³⁰ and its antagonist DWORF (34 amino acids),³¹ are involved in calcium regulation in skeletal muscles. Myoregulin and DWORF were also discovered in misannotated lncRNAs in rats and are important targets of muscle performance studies.^30,31

The polycistronic gene Enod40 exhibits two highly conserved smORFs in plants (12 and 24 amino acids in soybeans), which are expressed during early stages of root nodule organogenesis.¹⁴ Interestingly, these smORFs overlap, and their peptides bind to nodulin-100, a sucrose synthase subunit, suggesting their involvement in sucrose uptake control in nitrogen-fixing nodules.¹⁴

The transmembrane smORF peptide hemotin (88 amino acids) was identified in D. melanogaster. This peptide acts by regulating endosome maturation during phagocytosis, allowing the phagocytic digestion of microorganisms.¹²⁰ Hemotin deletion results in the accumulation of undigested material within endolysosomes, decreased resistance against infections and reduced lifespan.¹²⁰ Interestingly, the same study identified a homologous relationship between hemotin and stannin,¹²⁰ a vertebrate smORF peptide (88 amino acids in humans).¹²¹ Both smORF peptides exhibit sequence, structural and functional conservation.¹²⁰

Recently, the mitochondrial smORF peptide brawnin (71 amino acids length) was identified in vertebrates as an essential player in respiratory complex III assembly. In zebrafish, brawnin deletion causes complete complex III loss, which promotes early death.¹²²

6. smORFs as isoforms of major ORFs (isoformic smORFs)

Isoformic smORFs are small variants of major ORFs. The main mechanism of isoformic smORF generation is alternative splicing (Fig. 5),¹⁰ which is a widespread regulatory process throughout multicellular eukaryotes due to their greater numbers of introns.¹²³ Furthermore, other regulatory mechanisms, such as alternative transcription initiation (Fig. 5B), alternative polyadenylation (Fig. 5C) and alternative refCDS translation (Fig. 5D), could theoretically generate isoformic smORFs (reviewed in de Klerk and 't Hoen¹²⁴ and Touriol et al.¹²⁵) Even though smORF peptides translated from pseudogenes are not precise isoforms of their reference proteins, they may play a similar role as smaller forms of large CDS variants (Fig. 5E). Thus, isoform generation is an impressive mechanism underlying genetic variability,¹²⁶ and thousands of variants could fall within the smORF size thresholds.

Figure 5.

Mechanisms of isoformic smORF biosynthesis. Isoformic smORFs are generated via large ORF transcriptional editing, which fragments large CDSs into smaller variants that can fall within the smORF length limits. (A) Alternative splicing (AS) is the best described mechanism of isoformic smORF biosynthesis; however, other molecular processes, such as (B) alternative transcription initiation via alternative promoters, (C) alternative polyadenylation cleavage, (D) alternative refCDS translation mediated by downstream start codons and (E) the fragmentation of homologous pseudogenes, could theoretically generate isoformic smORFs. Black tracks represent exons; black lines represent introns; green lines represent ORF variants; red, yellow and blue circles indicate the respective processes on the left.

Isoformic smORF generation via alternative splicing is possible via different pathways, such as exon deletion, which promotes major ORF truncation to generate small fragment(s)¹²⁷ (Fig. 5A), and intron retention¹²⁸ (Fig. 5A), which theoretically inserts stop and/or start codons within a major ORF. These processes lead to the generation of smaller isoforms with different carboxy and amino termini, although premature termination codon (PTC) insertion can also trigger the nonsense-mediated mRNA decay pathway.¹²⁸ Moreover, more than one isoformic smORF can be generated by the same canonical RNA depending on the number of non-inactivating truncations produced.^129,130

6.1. Strategic insights into the investigation of isoformic smORFs

Vertebrates are potentially the best models for the discovery of isoformic smORFs regulated by alternative splicing in animals. Vertebrate species are thought to exhibit a higher occurrence of alternative splicing than representatives of non-chordate species.^130,131 Considering that the number of coding genes does not differ between invertebrates and vertebrates (e.g. ∼20,000 genes in humans <www.ncbi.nlm.nih.gov/genome/guide/human/> versus ∼20,000 genes in Caenorhabditis elegans¹³²) evolution, differentiation and complexity may not be precisely linked to the number of genes in a taxon but are instead linked to the diversity of variants produced.¹²³ Thus, many isoformic smORFs might be overlooked in vertebrates because alternative splicing (among other mechanisms) allows the number of transcripts to reach up to 10 times the number of precursor genes.¹³³ Importantly, the experimental prediction of alternative splicing variants is as challenging as the prediction of smORFs itself because splicing complexity has not been fully elucidated, especially in the case of deleterious mutants (reviewed in Blakeley et al.¹³⁴) however, a large-scale analysis in Physcomitrella patens (moss) identified 6,092 smORFs regulated by alternative splicing in 4,389 different genes.³⁹ The same study reported that isoformic smORF peptides tend to follow the activity pathways of their precursors as well as their amino acid usage,³⁹ which are important leads for functional investigations.

The bioinformatic prediction of isoformic smORFs performed only on transcriptome sequencing data is limited by de novo assemblies that could not discriminate alternative gene products without a reference genome. If genome sequences are available, the use of computational predictors of splicing sites,¹³⁵ splicing variants¹³⁶ or even alternative transcription initiation and polyadenylation sites¹³⁷ prior to smORF detection is advantageous, because these software can generate alternative transcription data as reference for transcriptome assemblers. On the other hand, the functional investigation of isoformic smORFs at the DNA level is particularly difficult due to the overlap between smORFs and their major variants. Thus, post-transcriptional and post-translational approaches, such as RNAi, in situ hybridization and antibody staining, as well as comparisons to ribosome profiling and shotgun proteomic data, are more suitable.

Pseudogene homologous analysis requires pseudogene prediction pipelines prior to smORF detection.¹³⁸ Then, an interesting strategy is the evaluation of evolutionary constraints on pseudogene homologous smORFs to detect coding potential for further investigation. Consistent with this strategy, 50 pseudogenes encoding homologous smORF peptides of canonical proteins were reported to undergo translation and the resulting peptides are under evolutionary constraints (Ka/Ks ratio < 0.3), which suggests potential functional roles.¹³⁹

smORF functional analysis of pseudogenes using CRISPR-Cas9 is also challenging. For instance, if the target pseudogene region is similar or identical to the parental gene, the specificity of CRISPR-Cas9 will be problematic because the guide RNA might misdirect the Cas9 nuclease to non-target regions.

6.2. Examples of isoformic smORF peptides

The most representative examples of isoformic smORFs encode interference peptides (small interfering peptides (siPEPs) or microproteins (miPs)) akin to microRNAs.³⁴ siPEPs/miPs consist of a single protein−protein interaction domain, which allows them to bind to larger proteins, typically transcription factors, generating non-functional complexes.^140,141 Moreover, siPEPs/miPs can control their targets via substrate competition.^140,141 siPEP/miP generation occurs via single-gene-unit expression or alternative splicing; in the second case, they can regulate their own alternative splicing transcript variants.^140–143

The tumour suppressor TEL (also known as ETV6) belongs to the ETS transcription factor family.¹⁷ Five alternative splicing variants of TEL may be generated by exon deletion.¹⁷ Two isoforms encode the smORF peptides TEL-c and TEL-d (both ranging from 20 to 30 amino acids), which lack the activity of their full-length variant suppressors but exhibit considerable expression in individuals with myelodysplastic syndrome-derived leukaemia.¹⁷ Interestingly, TEL-c and TEL-d exhibit lower expression in myelodysplastic syndrome before cancer progression, suggesting a link to smORF expression and leukaemia development.¹⁷

Members of the Phytochrome Interacting Factor (PIF) transcription factor family are important players in seed dormancy and germination control in response to environmental stimuli in A. thaliana.¹⁴⁴ A previous analysis showed that PIF6 contains an alternative splicing isoformic peptide of ∼180 amino acids known as PIF6-β, which is generated by the insertion of a premature stop codon and the excision of the DNA ligand domain due to the deletion of exon 3.¹⁴⁴ PIF6-β is highly expressed in seed development; however, PIF6 gene silencing, which includes full-length variant knockdown, increases primary seed dormancy. On the other hand, the overexpression of PIF6-β alone promotes dormancy reduction.¹⁴⁴ Even though the length of PIF6-β does not correspond to the most accepted smORF threshold of 100 codons, it is an interesting example of how alternative splicing can generate small variants with important biological effects. Unfortunately, annotation screenings have neglected the roles of alternative splicing smORF variants across species, possibly as a consequence of methodological obstacles in smORF detection.

7. smORFs as alternative CDSs in mRNAs

Alternative smORFs are alternatively or pervasively translated from the same mRNAs as canonical large CDSs. In contrast to isoformic smORFs, alternative smORFs are not variants of large CDSs generated via RNA editing but rather are different CDSs located in different parts of the mature mRNA.

In early genomics research, the ‘one gene, one protein’ dogma¹⁴⁵ was confronted by the capacity of mRNAs to encode more than one protein from alternative ORFs. In this context, the classic configuration of a eukaryotic mRNA was described as a monocistronic transcript divided into three regions: a CDS, usually consisting of a major ORF, flanked by 5′ and 3′ untranslated regions (UTRs) (Fig. 6A).¹⁴⁶

Figure 6.

Location and distribution of alternative smORFs. (A) Canonical monocistronic mRNA paradigm comprising a unique large CDS between untranslated regions (UTRs). (B) Alternative smORF division and distribution within an mRNA. Upstream smORFs are located in the 5′UTR, and their stop codons may extend across the reference CDS region. The start codons of overlapping smORFs strictly overlap with the reference CDS region, but their sequences may extend to the 3′UTR. Downstream smORFs are located totally within the 3′UTR. (C) Examples of well-known representative mRNAs exhibiting several alternative smORFs (sequence analysis by the authors). Alternative smORFs are commonly encountered in mRNAs, and their coding potential is still underappreciated. Yellow tracks represent reference CDSs (large ORFs); blue tracks represent alternative smORFs.

Mechanisms such as alternative splicing¹²⁶ and mature RNA editing^147,148 are well-known processes underlying alternative protein evolution from mRNAs; however, emerging evidence suggests that alternative ORF translation is also an important biological pathway contributing to genetic variability, either from overlapping genes^149–151 or alternative ORFs in unique transcripts.¹⁵² In the latter case, translation may involve post-transcriptional regulation mechanisms, such as the reinitiation of translation,¹⁵³ ribosomal frame shifting¹⁵⁴ and stop codon read-through.¹⁵⁵

Although de novo protein generation via alternative ORFs is atypical, especially in eukaryotes,^156,157 the distribution of coding alternative ORFs is speculated to be underestimated due to detection difficulties.¹⁵⁸ Proteomic and transcriptomic approaches corroborate the existence of putative alternative smORFs in the flanking regions of or overlapping with refCDSs in different frames,^{38,58,100,159} by definition comprising new polycistronic mRNAs.¹⁶⁰ For instance, recent data suggest that 15% of alternative smORFs in humans are preceded by Kozak consensus sequences,³⁸ suggesting that smORFs can show efficiency in ribosome recognition.^161,162 Importantly, hundreds of alternative smORFs with evidence of translation in humans are highly conserved in distant taxa, such as basal vertebrates, invertebrates and yeast.¹⁰⁰ Another study identified 149 alternative smORFs that are conserved among humans and rats. Dozens of these smORFs overlap with reference CDSs but do not show evolutionary sequence constraints on their codon composition.¹⁵⁹ Recently, a large-scale approach using cross-linked mass spectrometry followed by shotgun proteomics revealed that alternative ORFs act as regulators in NCH82 human glioma cell reprogramming via protein kinase A activation.¹⁶³

Alternative smORFs are an emerging frontier in the study of mechanisms of genetic variability³⁶; however, RNA-based phenotype analysis techniques such as RNAi and in situ hybridization cannot distinguish the roles of refCDSs and alternative smORFs since mRNAs are identical in these two cases. Thus, new tools such as CRISPR-Cas9 have arisen as a promising approach for studies on this topic, although modifications in overlapping smORFs still pose the challenge of refCDS constraints, which requires careful experimental planning.

Alternative smORFs present peculiarities inherent to their positions in transcripts. Alternative ORFs that are located in 5′UTRs, overlap with refCDSs or are located in 3′UTRs are referred to as upstream ORFs, overlapping smORFs and downstream smORFs, respectively ^8,35 (Fig. 6B). Alternative smORFs are frequently present in all types of mRNAs (some examples are presented in Fig. 6C), even though functions may be lacking. Thus, discovering which smORFs are biologically relevant is particularly challenging. The characteristics of each subgroup of alternative smORFs are detailed below.

7.1. smORFs as alternative CDSs in mRNAs: upstream ORFs

Upstream ORFs (uORFs) are smORFs located in 5′UTRs and have been reported in 20 − 50% of eukaryotic mRNAs.¹⁶⁴ uORFs play roles in several post-transcriptional control mechanisms, particularly by modulating ribosomal access to downstream refCDSs.¹⁶⁵ This mechanism promotes the regulation of translation efficiency and triggers mRNA degradation.¹⁶⁵ Post-transcriptional regulation promoted by uORFs is highly important to prevent the overexpression of central proteins of cellular networks; in this context, mRNAs of regulatory proteins such as transcription factors tend to evolve uORFs,^161,166 and mRNAs without uORFs encode more abundant proteins.^167,168 uORFs modulate the translation of several disease-related mRNAs (reviewed in Zhang et al.¹⁶⁹) including transcripts modulated during cancer.¹⁷⁰

uORF peptides can function by interacting with refCDS proteins or by binding to other molecules to trigger ribosomal stalling.^171–173 Cis-acting regulatory uORF peptides are known as peptoswitches, analogous to riboswitches, which are modulators of RNA transcription and translation.¹⁷⁴ On the other hand, the translation of uORFs can generate non-functional peptides, in which translation itself is the regulatory event.¹¹³ For instance, uORF stop codons can be recognized as PTCs, thereby triggering the nonsense-mediated mRNA decay (NMD) pathway, which promotes transcript degradation.¹⁷⁵ Another interesting regulatory mechanism occurs when uORF stop codons overlap with the initial portion of refCDSs (known as overlapping uORFs, or oORFs); in these cases, translational control is achieved via direct competition between the oORF and the refCDS for ribosomal coverage.¹⁷⁶

Even though uORFs perform several cis-regulatory roles, a significant portion may encode trans-acting peptides.^15,177–180

7.1.1. Strategic insights into the investigation of uORF peptides

Reannotation of mRNAs focussing on uORF discovery is a promising approach. For instance, in zebrafish, a computational analysis showed that over 60% of all protein-coding genes contain uORFs.¹⁸¹ This number can reportedly reach 50% in mammals^166,182,183 and 30% in plants¹⁸⁴; however, a small portion of uORFs show coding potential in terms of amino acid sequence conservation. Only 1.1% (87) of the uORFs found in mice, 1.7% (149) in humans, 0.3% (27) in zebrafish¹⁸¹ and 0.5% (44) in Arabidopsis¹⁸⁵ appear to be under selective pressure to maintain their encoded amino acid sequences. In zebrafish, over 60% of conserved uORFs show evidence of translation, as indicated by ribosome profiling.¹⁸¹ Interestingly, ∼20% of uORFs exhibit start codons in Kozak consensus optimal regions in humans.¹⁸⁶ In addition, some uORF peptides that are conserved between humans and mice are under stabilizing selection.¹³⁹ Based on the aforementioned data, we suggest that a conservative coding uORF annotation approach focuses on amino acid sequence conservation, the presence of Kozak consensus and stabilizing selection. In addition, the amino acid usage of uORFs is an important hint because it differs from that of canonical proteins,³⁴ and alternative start codons are common.^187,188 Considering that most functional uORFs perform cis-regulatory roles, sequences that do not follow these parameters are potentially non-coding or pervasively translated.

7.1.2. Examples of uORF peptides

uORF peptides have well-established roles in cis-acting regulatory pathways. For instance, the transcript of human S-adenosylmethionine decarboxylase, an enzyme involved in the responses to changes in polyamine levels, displays one of the smallest smORF peptides described to date (six amino acids) whose translation triggers ribosomal stalling and modulates the translation of its downstream refCDS.¹⁸⁹

Another interesting example comes from the model legume Medicago truncatula, in which the 62 amino acid peptide uORF1p is generated via partial intron retention in the mRNA of the transcription factor MtHAP2-1. uORF1p also regulates the translation of its refCDS via a trans-acting regulatory mechanism.¹⁹⁰ Once translated, uORF1p binds to the 5′ leader sequence of the MtHAP2-1 transcript, thereby decreasing MtHAP2-1 translation.¹⁹⁰

Another uORF peptide that acts via trans-acting mechanisms occurs in the 1A glucocorticoid receptor transcript.¹⁵ Receptor synthesis is completely inhibited by the deletion of the second (uORF-2) of five uORFs in its mRNA. Interestingly, uORF-2 translation results in a 93 amino acid peptide that also modulates the expression of the 1A glucocorticoid receptor via an unknown mechanism. Therefore, the uORF-2 peptide may be involved in the translation of this receptor, probably by interacting with other molecules.¹⁵

uORF peptides can also play roles that are wholly independent of the corresponding refCDS regulation. For instance, the gene encoding the MKKS protein, associated with McKusick−Kaufman syndrome, contains three regulatory uORFs. Two of these uORFs encode highly conserved peptides (63 and 45 amino acids in humans) that can be observed in the mitochondrial membrane, while the MKKS protein remains in the cytoplasm. These data suggest that uORFs can act independently from their respective refCDSs.¹⁷⁷

The 5′UTR of the c-akt proto-oncogene transcript displays an uORF encoding a 10 amino acid tumour rejection antigen (pRL1), which is recognized by cytotoxic T lymphocytes in BALB/c radiation-induced leukaemia RL♂1 cells.¹⁹¹ Interestingly, sequence analysis showed that pRL1 amino acid residues are identical to the 269 − 278 stretch of the protein expressed by the c-akt viral homologue v-akt.¹⁹¹ These findings demonstrate that uORFs can also play important immunological roles, although overlapping smORFs have been more frequently identified as antigen precursors.¹⁹²

7.2. smORFs as alternative CDSs in mRNAs: overlapping smORFs

Overlapping smORFs are smORFs that overlap with the refCDS in another frame. The start codons of overlapping ORFs strictly overlap with refCDSs and can extend to the 3′UTR,¹⁹³ unlike oORFs, as mentioned in the previous section. The great majority of overlapping ORFs are smORFs.^38,100 Several overlapping smORF peptides have been described, and their widespread expression suggests many different roles.^194–196

7.2.1. Strategic insights into the investigation of overlapping smORFs

Overlapping smORFs are abundant in mRNAs but were initially considered to represent a strategy for increasing genetic variability in size-restricted species genomes, such as those of prokaryotes^156,159,197; however, overlapping ORFs have also been discovered in complex eukaryotes.^{100,193,198,199} For instance, in humans, ∼41% of mRNAs contain at least one unannotated overlapping ORF, and most of these ORFs encode peptides smaller than 90 amino acids.¹⁰⁰ In another study, 217 overlapping ORFs containing Kozak consensus sequences were shown to be conserved in rodents and humans according to RefSeq transcript analysis.¹⁹³ To distinguish random sequences from coding stretches, some features must be considered. For example, overlapping smORFs suffer from the sequence constraints imposed by refCDSs because evolutionary pressures on one frame affect the others, thereby challenging phenotypic analysis; however, previous studies reported an evolutionary mechanism that allows overlapping ORF proteins to evolve less restrictively at both the sequence and structural levels.¹⁵⁷ For instance, studies analysing the CDSs of viral overlapping genes suggest that overlapping ORFs tend to encode structurally disordered proteins²⁰⁰ with codon-rich amino acids such as arginine, leucine and serine.²⁰¹ Another overlapping ORF feature is the oscillating amino acid modification rate, which differs from that of single-CDS genes.^202,203 Therefore, this mutation pattern can be used as an important parameter for overlapping smORF peptide detection.¹⁵⁹

7.2.2. Examples of overlapping smORF peptides

Several overlapping smORFs have been discovered and previously annotated within mRNAs (brief review in Andrews and Rothnagel⁸) In 1996, the gp75 melanoma antigen transcript that contains an overlapping smORF encoding a 24 amino acid peptide was identified as a tumour rejection antigen that is recognized by T-cells.¹⁶ Since then, several tumour antigens encoded by overlapping smORFs have been discovered,^204–210 indicating that overlapping smORFs are significant reservoirs of endogenous antigens.

Overlapping smORF antigens are commonly observed during viral infection or tumour cell growth.^192,211 A ribosome-based mechanism evolved to provide cryptically translated peptides that can be strictly used as substrates for antigen processing has been suggested.²¹² Moreover, this mechanism could represent one of the most significant pathways for the generation of endogenous antigens associated with the major class I histocompatibility complex (MHC class I).²¹³ Thus, this mechanism is an important topic of studies on immunological surveillance and new vaccine development.^214,215

Importantly, peptides with immunogenic potential are not unique products derived from overlapping smORFs. For example, the PRNP gene encodes the mammalian cellular prion protein (PrP),¹⁹⁶ a glycoprotein that is the causative agent of neurodegenerative diseases after deleterious structural folding.²¹⁶ The physiological functions of PrP are not entirely known because the PrP transcript is widely expressed not only in nerve tissues but also in the heart, skeletal muscle, intestine, uterus, and testis (reviewed in Sarnataro et al.²¹⁶). Interestingly, the diversity of the PrP transcript distribution may be associated with the highly conserved overlapping smORF that encodes the mammalian AltPrP peptide (73 amino acids). AltPrP is transported to mitochondria, and its stability is regulated by endoplasmic reticulum stress and proteasome inhibition.¹⁹⁶ Some of the toxic or protective functions attributed to PrP may actually be triggered by AltPrP.¹⁹⁶

The INK4a gene (also known as MTS1 or CDKN2) encodes a protein with tumour suppressor characteristics involved in cell cycle regulation called p16^INKa.²¹⁷ The INK4a transcript contains an overlapping smORF encoding a 132 amino acid peptide called p19ARF in humans. The ectopic expression of p19ARF in the fibroblast nucleus induces interphase G1 and G2 arrest, demonstrating that overlapping smORFs can also perform important roles in cell cycle control.²¹⁸

7.3. smORFs as alternative CDSs in mRNAs: downstream smORFs

Downstream smORFs are ORFs located in 3′UTRs. The coding potential of downstream smORFs is underestimated and poorly explored in comparison to that of other alternative smORFs³⁵; however, 3′UTRs contain important translational regulatory elements and subcellular localization signals, also contributing to eukaryotic transcript stability, tissue patterning processes, embryonic axis formation, mammalian spermatogenesis (reviewed in Wang et al.²¹⁹) and cancer.^170,220,221 In addition, 3′UTRs contain an SECIS (selenocysteine insertion sequence) element, a signal required for the insertion of the rare amino acid selenocysteine into UGA stop codons during the translation of canonical major ORFs.^222,223

A recent study showed that downstream smORFs also represent a widespread potential translation regulatory mechanism among vertebrates because the translation of downstream smORFs itself is required for the increased translation of reference major ORFs, depending on the number of downstream smORFs in the mRNA. Importantly, the amino acid sequence and smORF peptide length do not influence this regulatory mechanism²²⁴; however, downstream smORF peptides have been scarcely reported in the literature.^100,225 Although there is significant evidence of ribosomal coverage in 3′UTRs,^226,227 these events are often associated with delays in ribosome decoupling after translation or stop codon read-through,¹⁵⁵ where the ribosome does not recognize the refCDS stop codon and proceeds in the reading of the entire 3′UTR.²²⁸

7.3.1. Strategic insights into the investigation of downstream smORFs

The prediction of coding downstream smORFs is particularly challenging. Although the translation of downstream smORFs has been described,^{37,100,139,188} most of the previously reported examples exhibit a low translation efficiency in ribosome profiling analysis; however, some downstream smORFs are more highly translated,¹³⁹ which suggests that rare coding downstream smORFs exist and await annotation.

Interestingly, an important phenomenon has been described wherein 3′UTRs are cleaved by an unknown post-transcriptional mechanism, but polyadenylation sites and downstream smORFs are retained in a new independent transcript.²²⁹ Importantly, 3′UTR transcripts usually exhibit different expression patterns than their parental mRNAs,²³⁰ as described for up to 50% of 3′UTR transcripts in rats.²²⁹ RNAs generated by 3′UTR cleavage have evolved non-coding activities,²³¹ and traditional gene prediction approaches consider their coding capacity to be low, classifying them as potential ncRNAs²²⁹; however, the coding capacity of 3′UTR transcripts remains unclear because coding smORFs are usually dismissed by traditional gene prediction methods. Thus, studies designed to explore the coding potential of smORFs in 3′UTR transcripts could be promising, but the mapping of 3′UTR transcripts is difficult due to the low availability of 3′UTR read annotation, coverage and assembly data in public repositories, even for well-studied CDS transcripts.²³²

7.3.2. Examples of downstream smORF peptides

The first coding downstream smORF was identified in the H60 histocompatibility gene, encoding the eight amino acid antigen LYL8, which is presented by MHC class I to the immune system, thereby suppressing cytotoxic T-cell activation and inducing immunological self-tolerance against endogenous polypeptides.^225,233 Interestingly, the observation of the insertion of stop codons between the refCDS and downstream smORF as well as the alternation of frames showed that LYL8 translation does not occur via stop codon read-through because LYL8 bioactivity remained unchanged.^225,234 These data suggest that downstream smORF translation is possible via direct ribosome recognition, independent of the refCDS.

Other findings have shown that stop codon read-through is an important mechanism for downstream smORF translation. Interesting findings were obtained from studies involving aminoglycoside antibiotics, which increase translational frequency via premature termination codon (PTC) read-through and are indicated for use against disorders caused by defective proteins generated due to PTCs.²³⁵ Cells treated with the aminoglycoside gentamicin undergo an apparent autoimmune response involving the translation of downstream smORF antigens that are able to activate CD8+ T cells.²³⁵

The MRVI1 (murine retrovirus integration-site 1) gene encodes a protein with myeloid leukaemia suppressor activity.²³⁶ MRVI1 also encodes a downstream smORF peptide (∼95 amino acids) that interacts with the tumour suppressor BRCA1 (BReast CAncer type 1 susceptibility protein) in the HeLa cell nucleus.¹⁰⁰ The MRVI1 downstream smORF peptide was possibly previously rejected as a bioactive product due to its out-of-frame position with respect to the refCDS.¹⁰⁰ These data suggest that other functional downstream smORFs may have been dismissed for the same reason in large-scale analyses.

8. Conclusion and future prospects

The integration between omic sciences and bioinformatics has enabled the study of previously obscured topics in systems biology, such as junk DNA, wherein hundreds of smORFs have been discovered as new players in developmental biology, cancer, neuropathologies, transcription/translation control, tissue physiology, immunological surveillance, and responses to environmental stimuli, among other phenomena.

The emergence of smORF peptides as a significant and non-sporadic phenomenon occurred during mid-2010, when the NGS era allowed deep sequencing analysis. Since then, numerous smORFs have been discovered, especially during the last 5 years, when hundreds of lncRNAs were reannotated as smORF transcripts encoding important and essential functional peptides. Furthermore, housekeeping RNAs such as rRNAs, pri-miRNAs and circRNAs have been identified as coding smORF reservoirs.

In summary, the new smORF classification presented here will help to direct the further development of bioinformatics and functional studies. Genome annotation screenings have dismissed the coding potential of smORFs owing to the lack of knowledge about this new class of genes; thus, this review will help researchers uncover these important elements of molecular biology by offering many insights into the study of each smORF class. The functional characterization of even more smORF peptides in the future will provide evidence of the number of essential smORFs in different taxa, supporting comparative analysis. Future studies should also address the evolutionary trends of smORFs among different phyla, such as whether the classes described here differ among species. Studies comparing smORFs at the base of metazoan and non-bilaterian groups, such as sponges, cnidarians, placozoans and ctenophores, would be particularly interesting. Additionally, a comparison of the roles and conservation of smORFs during speciation and/or whole-genome duplication events might provide new insights into the origin and function of this interesting class of genes. Finally, the recent discussion about non-coding smORFs as precursors of coding gene birth^32,33 is one of the new frontiers of evolutionary biology, which may lead to a whole new area of future research. Hence, understanding smORF diversity and its singularities is essential to discover evolutionary innovations in this hidden coding DNA world.

Authors’ contributions

D.G.-A., D.A.T. and R.N.-da-F. contributed equally to the writing of this manuscript.

Funding

D.G.-A. was a master’s student of Federal University of Rio de Janeiro with scholarship financed by Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES. D.A.T. and R.N.-da-F. were supported by Fundação de Amparo à Pesquisa do Estado do Rio de Janeiro - FAPERJ (D.A.T.: E-26/202.736/2019; R.N.-da-F.: E-26/211.169/2019, E-26/202.605/2019, E-210.264/2018) and Conselho Nacional de Desenvolvimento Científico e Tecnológico - CNPq (grant numbers unavailable due to system lockdown)

Conflict of interest

None declared.